Automatic detection and sizing of pores from radiographic images using the Hough transform

Abstract

We use a pattern recognition approach to assist the human interpreter in evaluating the quality of welds from radiographic images, especially tube-to-tubesheet weld joints in the steam generator assembly of the prototype fast breeder reactor (PFBR), that require stringent inspection as the joints separate water and steam. In this paper, the circular Hough transform is used for automatic detection of circular discontinuities that are present in the weld joints, and we use morphological parameters to classify these defects in terms of their size. Using this approach, a success rate of 99% detection of circular discontinuities is achieved, by segmenting the image into smaller regions of interest and using the circular Hough transform along with morphological parameters.